---
title: "The Seriality Gap in Video Diffusion Models"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.13031v1"
permalink: "https://feed.astralor.com/items/itm_132dd57a5b5d60298c4cac7e"
published: "2026-07-14T17:59:22.000Z"
score: 79
channel: "ai"
tags: ["arXiv","Causal Chain","Serial Reasoning","Hard-sphere dynamics","Autoregressive Generation","Video Diffusion Models"]
---

# The Seriality Gap in Video Diffusion Models

## AI commentary

> This paper does something I've been waiting for: stress-test video diffusion models on tasks requiring step-by-step causal reasoning. Unsurprisingly, models show a clear structural weakness. For researchers in video generation, world models, or physics simulation, this is a must-read—it highlights the ceiling of the current paradigm and hints at a return to autoregressive or hybrid architectures.

## At a glance

This arXiv paper, through controlled experiments on multi-ball hard-sphere dynamics, reveals that standard bidirectional video diffusion models degrade significantly as causal chains lengthen, even with more denoising steps. Control experiments with single-ball dynamics isolate the cause as dependent-event structure rather than video length. Intervention studies show that methods increasing effective serial computation—autoregressive generation, blockwise generation, and architectural depth—disproportionately improve performance. The authors identify this as the 'seriality gap': a mismatch between tasks requiring growing serial computation and video diffusion models whose denoising loop does not provide scalable serial computation. They further prove that, for deterministic video prediction, denoising steps do not add serial computation beyond the backbone, indicating a structural obstacle for video diffusion on serial reasoning and simulation tasks.

## Key points

- Standard bidirectional video diffusion models degrade as causal chains lengthen, and adding denoising steps does not help.
- The root cause is the serial computation requirement from event dependencies, not video length.
- Methods that increase serial computation—autoregressive generation, blockwise generation, and deeper networks—disproportionately improve performance.
- The paper theoretically proves that denoising steps do not add serial computation beyond the backbone, creating a structural obstacle.

## Full text

When one ball strikes another, then another, video models should predict the consequences of each bounce. In controlled experiments on multi-ball hard-sphere dynamics, we find that the performance of standard bidirectional video diffusion degrades as the causal chain lengthens, even when provided more denoising steps. In a length-matched single-ball control, where ball-ball interactions are absent, the degradation largely disappears, isolating dependent-event structure rather than video length as the cause. Across intervention studies, methods that increase effective serial computation improve performance disproportionately, including autoregressive/blockwise generation and architectural depth. We identify this pattern as the seriality gap: a mismatch between tasks requiring growing serial computation and video diffusion models whose denoising loop does not provide scalable serial compute. We then prove that, for deterministic video prediction, denoising steps do not add serial computation beyond the backbone, indicating a structural obstacle for video diffusion on serial reasoning and simulation tasks.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_132dd57a5b5d60298c4cac7e